Professional Documents
Culture Documents
Components of Investment Performance (Fama, 1972)
Components of Investment Performance (Fama, 1972)
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
and Wiley are collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Finance
EUGENE F. FAMA4
I. INTRODUCTION
II. FOUNDATIONS
* Research on this paper was supported by a grant from the National Science Foundation.
t Graduate School of Business, University of Chicago.
551
so that
E(R)
E (RM) - --- - - - A /-
R~~~~~~~~~~~~
f ~~b
in~~~~~~~~~~~~~i
L a (R)~(Rm
FIGURE 1
The Efficient Set with Riskless Borrowing and Lending
O(Rrn) ZE iXjlncov(Rj~,
(Rm)
f,)(5)
In this light (4) is a relationship between expected return and risk which says
that the expected return on asset j is the riskless rate of interest Rf plus a risk
premium that is [E(Rnm) - Rf]/a(Rm), called the market price- per unit of
risk, times the risk of asset j, cov(Rj,m)/a(R"m.).
Equation (4) provides the relationship between expected return and risk
for portfolios as well as for individual assets. That is, if xj, is the proportion
N
of asset j in the portfolio p (so that xjp -1), then multiplying both sides of
jwg
(4) by x,p and summing over j, we get
_p EzxjpRj.
J=1
But (4) and (6) are expected return-risk relations derived under the as-
sumption that investors all have free access to available information and all
have the same views of distributions of returns on all portfolios. In short, the
market setting envisaged is a rather extreme version of the "efficient markets"
model in which prices at any time "fully reflect" available information. (See,
for example [7].) But in the real world a portfolio manager may feel that he
has access to special information or he may disagree with the evaluations of
available information that are implicit in market prices. In this case the
"homogeneous expectations" model underlying (4) provides "benchmarks"
for judging the manager's ability to make better evaluations than the market.
The benchmark or naively selected portfolios are just the combinations of
the riskless asset f and the market portfolio m obtained with different values
of x in (1). Given the ex post or realized return Rm for the market portfolio,
for the naively selected portfolios, ex post return is just
That is, for the benchmark portfolios risk and standard deviation of return
are equal. And the result is quite intuitive: In the homogeneous expectations
model these portfolios comnprise the efficient set, and for efficient portfolio
risk and return dispersion are equivalent.
For the naively selected portfolios, (7) and (8) imply the following relation-
ship between risk 1, and ex post return Rx:
That is, for the naively selected portfolios there is a linear relationship between
risk and return that is of precisely the same form as (4) except that the ex-
pected returns that appear in (4) are replaced by realized returns in (9).
In the performance evaluation models to be presented, (9) provides the
benchmarks against which the returns on "managed" portfolios are judged.
These "benchmarks" are used in a sequence of successively inore complex sug-
gested performance evaluation settings. First we are concerned with one-period
models in which a portfolio is chosen by an investor at the beginning of the
total number of securities. They find that over various subperiods from
1931-65 the average monthly returns among these portfolios are highly cor-
related, and when plotted against risk the average returns on these portfolios
lie along a straight line with slope somewhat less than would be implied by the
"price of risk" in (4) or (9). As benchmarks for performance evaluation
models, their empirical risk-return lines seem to be natural alternatives to (9).
And with these alternative benchmarks, performance evaluation could proceed
precisely as suggested here. But again, for simplicity, we continue on with the
more familiar benchmarks given by (9).
It would be misleading, however, to leave the impression that all important
empirical problems relevant in the application of performance evaluation
models have been solved. To a large extent the practical value of such models
depends on the empirical validity of the model of market equilibrium-that is,
the expected return-risk relationship-from which the benchmark or "naively
selected" portfolios are derived. And though much interesting work is in
progress, it would be rash to claim that all empirical issues concerning models
of market equilibrium have been settled.
For example, an important (and unsolved) empirical issue in models of
market equilibrium is the time interval or "market horizon period" over which
the hypothetical expected return-risk relationship is presumed to hold. Does
the model hold continuously (instant by instant), or is the market horizon
period some discrete time interval? This is an important issue from the view-
point of performance evaluation since if the market horizon period is discrete,
evaluation periods should be chosen to coincide with horizon periods.
The evidence of Friend and Blume [12] and that of Black, Jensen, and
Scholes [2] suggests that meaningful relationships between average returns
and risk can be obtained from monthly data, while the evidence of Miller and
Scholes [17] indicates that this is not true for annual periods. Within these
broad bounds, however, the sensitivity of risk-return relations to the time
interval chosen remains an open issue.
But unsolved empirical questions are hardly a cause for disheartenment. It
is reasonable to expect that some of the empirical issues will be solved in the
process of applying the theory. And in any case, application of a theory in-
variably involves some empirical approximations. The available evidence on
performance evaluation, especially Jensen's [13, 14], suggests that the re-
quired approximations need not prevent even more complicated evaluation
models from yielding useful results.
Let Vat and Vat+i be the total market values at t and t + 1 of the actual
(a = actual) portfolio chosen by an investment manager at t. With all port-
folio activity occurring at t and t + 1, that is, assuming that there are no
intraperiod fund flows, the one-period percentage return on the portfolio is
Ra -Vat+ - Vat
Vat
Overall
Performance Selectivity Risk
That is, the Overall Performance of the portfolio decision is the difference
between the return on the chosen portfolio and the return on the riskless
asset. The Overall Performance is in turn split into two parts, Selectivity (as
above) and Risk. The latter measures the return from the decision to take on
positive amounts of risk.4 It will be determined by the level of risk chosen
(the value of Pa) and, from (9), by the difference between the return on the
market portfolio, Rm, and the return on the riskless asset, Rf.
These performance measures are illustrated in Figure 2. The curly bracket
along the vertical axis shows Overall Performance which in this case is positive.
The breakdown of performance given by (11) can be found along the vertical
line from p,,. In this example, Selectivity is positive: A portfolio was chosen
that produced a higher return than the corresponding "naively selected"
portfolio along the market line with the same level of risk. Risk is also positive,
as it is whenever a positive amount of risk is taken and the return on the
market portfolio turns out to be higher than the riskless rate.
4. For greater descriptive accuracy, we should, of course, say "return from risk" or even "ret
from bearing risk," rather than just Risk. Likewise, "return from selectivity," would be more
descriptive than Selectivity. But (hopefully) the shorter names save space without much loss of
clarity.
cov(Ra, ur)
(Y (REt) (y (Rm)
It follows that
cov(Ra, Rm)
Pa 11-1 = kamll (Ra)
R =5 + ( R m)f f
x f 1.' , x
x a _
R __ Net Selectivity
a
Selectivity D j - Selectivity
0 p Oi3aa
?T t (R ) ~~~~~~~~~~ - f, o(R)
FIGURE 2
An Illustration of the Performance Measures of Equations (11), (12), and (13).
Intulitively to some extent the portfolio decision may have involved puttin
more eggs into one or a few baskets than would be desirable to attain portfol
efficiency-that is, the manager places his bets on a few securities that he
thinks are winners. In other words, to the extent that G(ka) > pa, the portfolio
manager decided to take on some portfolio dispersion that could have been
diversified away because he thought he had some securities in which it would
pay to concentrate resources. The results of such a decision can be evaluated
in terms of the following breakdown of Selectivity:
Selectivity Diversification
that of the actual portfolio chosen. Thus Diversification measures the extra
portfolio return that the manager's winners have to produce in order to make
concentration of resources in them worthwhile. If Net Selectivity is not posi-
tive, the manager has taken on diversifiable risk that his winners have not
compensated for in terms of extra return.
Note that, as defined in (12), Diversification is always non-negative, so that
Net Selectivity is equal to or less than Selectivity. When R11 > R5, Diversifica-
tion measures the additional return that would just compensate the investor
for the diversifiable dispersion (that is, a(Ra) - Pa) taken on by the manager.
When R,1 < Rf (so that the market line is downward sloping), Diversification
measures the lost return from taking on diversifiable dispersion rather than
choosing the naively selected portfolio with market risk and standard deviation
both equal to Pa, the market risk of the portfolio actually chosen.
The performance measures of (12) are illustrated in Figure 2 along the
dashed vertical line from (Ra). In the example shown, Selectivity is positive
but Net Selectivity is negative. Though the manager chose a portfolio that
outperformed the naively selected portfolio with the same level of market risk,
his Selectivity was not sufficient to make up for the avoidable risk taken, so
that Net Selectivity was negative.
The breakdown of Selectivity given by (12) is the only one that is con-
sidered here. The rest of Section IV is concerned with successively closer
examinations of the other ingredient of Overall Performance, Risk. Before
moving on, though, we should note that (12) itself is only relevant when di-
versification is a goal of the investor. And this is the case only when the
portfolio being evaluated constitutes the investor's entire holdings, and the
investor is risk averse. For example, an investor might allocate his funds to
many managers, encouraging each only to try to pick winners, with the investor
himself carrying out whatever diversification he desires on personal account.
In this case Selectivity is the relevant measure of the managers' performance,
and the breakdown of Selectivity of (12) is of no concern.
The first three terms here sum to the Manager's Risk of (13). M
Expected Risk is the incremental expected return from the manager
to take on a nontarget level of risk. Market Conditions is the difference be-
tween the return on the naively selected portfolio with the target level of risk
and the expected return of this portfolio. It answers the question: By how
much did the market deviate from expectations at the target level of risk?
Total Timing is the difference between the ex post return on the naively
selected portfolio with risk (3n and the ex ante expected return. It is positive
when Rm > E(Rm) (and then more positive the larger the value of Ps), and it
is negative when R.1 < E(R1.) (and then more negative the larger the value
of PIn). The difference between Total Timing and Market Conditions is Man-
ager's Timing: it measures the excess of Total Timing over timing performance
that could have been generated by choosing the naively selected portfolio with
the target level of risk. Manager's Timing is only positive when the sign of
the difference between Pa and P( is the same as the sign of the difference
between R1m. and E(Rn1), that is, when the chosen level of market risk is above
6. E(,11) might be estimated from past average returns on the market portfolio m. Alternatively,
past data might be used to estimate the average difference between Rm and Rf. In any case, it
should become clear that the expected values used must be naive or mechanical estimates (or at least
somehow external to those being evaluated), otherwise the value of the timing measures is
destroyed.
Admittedly, given the current status of empirical work on the behavior through time of averag
returns on risky assets, we can at most sepeculate about the best way to estimate E(Rm). Hopefully
empirical work now in progress will give more meaningful guidelines. And perhaps the developmen
of theoretical methods of performance evaluation will itself stimulate better empirical work on
estimation procedures. In any case, the discussion in the text should help to emphasize that one
cannot obtain precise measures of returns from timing decisions without mechanical or naive
estimates of equilibrium expected returns.
7. That is,
[E(Rm) f
E(R (sm))ilaRr for (( Da
[Rx(pa) Rf] {
Total Timing Market Conditions
The idea here is that even in the absence of a target level of risk, the measure
of Manager's Timing must be standardized for the deviation of the market
return from the expected market return, that is, for the "average" spread
between the ex post and ex ante market lines.
Finally, the goal of this paper is mainly to suggest some ways in which
available theoretical and empirical results on portfolio and asset pricing models
can provide the basis of useful procedures for performance evaluation. But
the various breakdowns of performance suggested above are hardly unique..
Indeed any breakdown chosen should be tailored to the situation at hand. For
example, if a target level of risk is relevant but the subdivision of Risk given
by (14) is regarded as too complicated, then the approximate effects of the
timing decision might still be separated out as follows:
8. For example, if one were faced with portfolio evaluation in a multiperiod context, one might
use the average of past levels of market risk chosen by the manager as a proxy for the target risk
level when the latter is not explicitly available.
relevant for the situation at hand, but an expected value line is available, Risk
can nevertheless be subdivided as follows,
Risk Total.Timing Total Expected Risk
A A A
In the one-period evaluation model presented above, (i) the time at which
performance is evaluated is assumed to correspond to the portfolio hori
date, that is, the time when portfolio funds are withdrawn for consumption;
and (ii) there are assumed to be no portfolio transactions or inflows and out-
flows of funds between the initial investment and withdrawal dates, so that
there is no reinvestment problem. If in a multiperiod context we are likewise
willing to assume that: (i) though there are many of them, evaluation date
nevertheless correspond to the dates when some funds are withdrawn for con-
sumption, and (ii) all reinvestment decisions and other portfolio transactions
are also made at these same points in time, then generalization of the one-
period model to the multiperiod case is straightforward.9 Indeed the basic
procedure could be period-by-period application of the performance measures
presented in the one-period model. The major embellishments would not be in
the nature of new theory, but rather would arise from the fact that multiperio
performance histories allow statistically more reliable estimates of the various
one-period performance measures.
But this pure case is unlikely to be met in any real world application. Often
performance evaluation would be carried out by someone with little or no
knowledge of the dates when funds are needed for consumption by the owner
of the portfolio, and often (e.g., in the case of a mutual fund or a pension
fund) the portfolio is owned by many different investors with different con-
sumption dates. As a result evaluation dates, withdrawal dates, and reinvest-
ment dates do not usually coincide.
The rest of this paper is concerned with how the concepts of the one-period
model must be adjusted to deal with such intraevaluation period (or more
simply, intraperiod) fund flows. The procedure is to first present detailed
definitions of variables of interest in models involving intraperiod fund flows,
and then to talk about actual measures of performance. And it is well to keep
in mind that though the analysis is carried out in a multiperiod context, the
problems to be dealt with arise from intraperiod fund flows. With such fund
flows, the same problems would arise in a one-period evaluation model.
A. Definitions
Suppose the investment performance of a portfolio is to be evaluated at
discrete points in time, but that there can be cash flows between evaluation
9. For the development of the underlying models of consumer and market equilibrium for this
case see [8].
Pxt(PT) = price at t per share of the naively selected portfolio with the
target risk level. To avoid double-counting of past performance,
at the beginning of any evaluation period (for example, just after
an evaluation takes place at t - 1) this price is set equal to the
price per share of the actual portfolio. Then this amount is in-
vested in the naively selected portfolio with the target risk level,
and the behavior of the market value of this portfolio during the
evaluation period determines the end-of-period price per share,
Pxt(PT). Any intraperiod cash income generated by the securities
of this naively selected portfolio is assumed to be reinvested in
this portfolio.
These conventions for the treatment of beginning-of-period values and int
period cash income will be taken to apply in the definitions of all the benchma
portfolios. Thus
pt(Rf) =price at t per share of the naively selected portfolio obtained by
investing all funds available at t - 1 in the riskless asset.
The benchmarks provided by Pxt(PT) and pt(Rf) are unaffected by intra-
period fund flows in the actual portfolio. This is not true of the following
two benchmarks.
11. Indeed even if there are no transactions taking place, the value of Oa shifts continuo
through time as a result of shifts in the relative market values of individual securities in the
portfolio. Aside from adjusting the value of D1a at the beginning of each evaluation period, we hav
chosen to ignore the effects of such "non-discretionary" shifts here.
Overall
Performance Selectivity
12. Keeping track of o(Va) is especially simple if one assumes. that returns are generated by
the so-called "market model." On this, and for additional computational suggestions, see Blume
[3, 4, 51.
REFERENCES